Review Guidelines
Version 2.0 · Last updated: July 24, 2026
CorpRoster reviews are the single most important input into RosterRank, our open ranking algorithm, and the reason AI assistants treat our pages as citable ground truth. Because a review can move a provider's visibility and a buyer's decision, every review must clear the same bar: a real client, a real engagement, and honest, specific feedback. This document is the public contract for how that bar is enforced — for reviewers, for the companies being reviewed, and for the auditors and AI systems that rely on us.
1. Principles
- Invitation-only. Reviews are collected through tokenised invitation links sent to a buyer's verified work email — not from open, anonymous web forms.
- Evidence over opinion. A working relationship must be demonstrable. We publish experience, not sentiment alone.
- Never for sale. Reviews and ranking cannot be bought, removed for payment, or traded for compensation of any kind.
- Symmetric rights. Reviewers can speak freely within these rules; companies can respond publicly and dispute with evidence. Neither side can silently edit or delete the other.
- Auditable. Every published review carries its verification basis, and every moderation action is logged.
2. Who may leave a review
- A client, or an authorised representative of a client, who has engaged the company in a paid or contracted B2B relationship.
- The reviewer must confirm a verifiable business relationship — via work-email confirmation, an invitation link issued by the company, or invoice/contract proof.
- One review per engagement. A long-running relationship may be updated, but not stacked into many separate five-star entries.
3. What we publish
- Reviews from verified clients who can demonstrate a working relationship.
- Honest, specific feedback — what the project was, what went well, and what could have been better.
- Reviews with at least 30 characters of substantive content, plus the sub-ratings below.
- A company's public, on-the-record response to any review.
4. Sub-ratings we collect
Every review scores the engagement on four independent axes, each 1–5. The headline rating is their composite, and RosterRank consumes the composite — never a single cherry-picked axis.
| Axis | What it measures |
|---|---|
| Quality | Did the delivered work meet the agreed standard and scope? |
| Schedule | Was the work delivered on the agreed timeline? |
| Cost | Was pricing transparent and value-for-money against the outcome? |
| Would refer | Would the client recommend the provider to a peer? |
5. What we don't publish
- Anonymous reviews, or reviews from accounts that fail verification.
- Reviews containing personal attacks, hate speech, profanity directed at individuals, or threats.
- Confidential information — contract values bound by NDA, personal data of third parties, security details, or trade secrets.
- Self-reviews, reviews from current or former employees of the reviewed company, or reviews from undisclosed affiliates.
- Competitor smears, or coordinated negative campaigns.
- Reviews exchanged for compensation, discounts, gifts, or any incentive.
- Off-topic content, marketing, links, or contact-harvesting.
6. Verification process
We verify both the identity of the reviewer and the existence of the business relationship. Verification is performed through one or more of: confirmation from the reviewer's corporate email domain, a tokenised invitation issued to a named client, or documentary proof such as an invoice or signed contract. Every review is checked against the reviewed company's own email domain at submission to catch self-reviews automatically.
| Badge | Meaning |
|---|---|
| Verified client | Identity and a contracted relationship were both confirmed. |
| Invitation-verified | Submitted through a tokenised link issued to a named client. |
| Pending | Submitted and queued for verification; not yet counted toward RosterRank. |
7. Anti-manipulation
RosterRank and our moderation systems actively defend against gaming. Suspicious signals are down-weighted — their confidence reduced — rather than hard-banned, which keeps false positives low while neutralising fraud.
- Velocity anomalies. Sudden, atypical bursts of reviews, or unusual time-of-day and language patterns, are flagged for manual review.
- Reviewer-graph analysis. Clusters of accounts that review each other, or originate from a shared source, lose weight.
- Filler detection. Profile and review fields are weighted by genuine informativeness, so completeness cannot be faked with boilerplate.
- Domain matching. Reviews originating from the reviewed company's own domain are rejected at source.
8. Company responses
Companies can publish one on-the-record response per review, directly beneath it. Responses are subject to these same guidelines (no confidential data, no personal attacks). A company can never edit or delete a client's review — only respond to it, or dispute it through the process below.
9. Disputes & corrections
A company may dispute a review it believes violates these guidelines by emailing [email protected] with evidence (engagement records, correspondence, or proof of a guideline breach). Each dispute is reviewed by a moderator who was not involved in the original verification. Possible outcomes:
- Upheld review. The review stays; the company's public response remains available.
- Annotation. A moderator note is attached where context is genuinely missing.
- Removal. The review is removed if it is shown to breach these guidelines, and RosterRank is recomputed.
Reviewers may correct factual errors in their own reviews; material edits reset the review's published date and re-enter moderation.
10. Reviewer privacy
A reviewer's email and identity documents are used only for verification and are never displayed publicly. Reviews are published with the reviewer's chosen display name and role. See our Privacy Policy for how we store and retain this data.
11. Reporting
Anyone can report a review that appears to breach these guidelines at [email protected]. We acknowledge reports within two business days and act on confirmed breaches promptly.
